search

JuliaImageRecon / RegularizedLeastSquares.jl

MIT License
20 stars 9 forks source link

increase accuracy of power iterations and remove flag `normalize_rho` #86

Closed JakobAsslaender closed 2 months ago

JakobAsslaender commented 2 months ago

Hi @tknopp,

I increased the default accuracy of the power iterations to ensure the correction step size of FISTA, optISTA, POGM reconstuctions.

I also changed the interface by removing the normalize_rho flag and building the power iterations into the default call of rho, increasing the flexibility to modify the settings of the power iterations as needed.

Any chance you could give me a quick thumbs up on this, I would like to go ahead and release this code, as we want to start a bunch of recons with it, and I would prefer not doing this on branch that less permanence than are release.

Many thanks and have a great weekend!

nHackel commented 2 months ago

As far as I can see the failing CI is just buildkite. That is CI that runs on GPUs, which wouldnt work for the master anyway atm.

nHackel commented 2 months ago

Ah and it fails because the master branch doesnt have the config yet.

From a CI standpoint you can feel free ro merge. If this is merged to the master I can also do the merge with the gpuExtension branch. Or do you want to do GPU recos with this?

JakobAsslaender commented 2 months ago

Right now we want to run recons on the CPU, but we are working on the GPU implementation in parallel, so yes, we do.

Ok, then I am just waiting for either @tknopp or you to give me a thumbs up content-wise.

tknopp commented 2 months ago

Looks good to me!